Re: [Bug-gnulib] getline bug?

[P]

Paul Eggert wrote:
> Jim Meyering <address@hidden> writes:
>
> > Simon's solution (of changing the default to be `don't use unlocked-io')
> > is fine with me.
>
> OK, I installed the following into gnulib, so that the change is done
> systematically for all gnulib modules.  I also installed the
> coreutils-relevant part of this patch into coreutils.  There are one
> or two other coreutils-only changes but I'll note them only on
> bug-coreutils.

This got me thinking of work I previously did in this area.
I actually prompted the use of unlocked IO stuff here:
http://www.mail-archive.com/address@hidden/msg00531.html

In general I don't agree with the unlocked versions of the glibc
functions being exposed. Couldn't glibc set a flag whether to
use locking or not when pthread_create is called?
See the inlined message here: 
http://marc.theaimsgroup.com/?l=linux-kernel&m=103251587403711&w=2
Note it's not feasible to do at compile time as indictated here:
http://www.mail-archive.com/address@hidden/msg02537.html

Also I really think that coreutils/lib/linebuffer.c should use getline
when available, as it's faster than even unlocked IO for any lines
over 2 characters (and only slightly slower in that case).
http://www.pixelbeat.org/linebuffer-2.0.21-getline.diff

Pádraig.